Computer Chess 2004

1
Computer Chess 2004

• How far have we come?
• By Mike Donner
• And
• Dan Mathia

2
Goals

• To see how computer chess has evolved throughout
  the years
• To show some of the theory behind the Computer
  Chess Masters
• To demonstrate the results of different chess
  programs squaring off for dominance of our
  Artificial Chess world

3
History
1957 - Newell and Simon predicted that a computer
would be chess champion within ten years Simon
I was a little far-sighted with chess, but there
was no way to do it with machines that were as
slow as the ones way back then 1958 - First
computer to play chess was an IBM 704 - about one
millionth capacity of deep blue. 1967 Mac Hack
competed successfully in human tournaments 1983
Belle obtained expert status from the United
States Chess Federation Mid 80s Scientists at
Carnegie Mellon University started work on what
was to become Deep Blue. Project moved to IBM in
1989
4
Computer Chess Masters

• Players we did not have access but will discuss
  to
• Deep Thought 1987
• Deep Blue 1997
• Hal (2001 Space Odyssey)
• Others between 1950s to1980s
• Mac Hack 3
• Chess 3
• Chess 4
• Belle

5
Deep Thought

• Built at Carnegie-Mellon University in the 1980s
• Example Deep Thought Code
• define min(x,y) ((x) lt (y) ? (x) (y))
• define max(x,y) ((x) gt (y) ? (x) (y))
• How Deep Thought thought
• Used different table for evaluation function
• Piece placement table
• Wieghts were different for each side.
• Stong black move weak white move
• Piece value table
• Queen 9
• Rook 5
• Knight 3
• Pawn 1

6
Deep Blue

• First Created in 1997 (Deep thought All grown up)
• algorithm
• iterative-deepening alpha-beta search,
  transposition table, databases including openings
  of grandmaster games (700,000), and endgames (all
  with 5 pieces or more pieces remaining)
• hardware
• 30 IBM RS/6000 processors
• They do high level decision making
• 480 custom chess processors
• all running in parallel
• They do
• deep searches into the trees
• move generation and ordering,
• position evaluation (over 8000 evaluation points)
• average performance
• 126 million nodes/sec., 30 billion position/move
  generated, search depth 14

7
Hal

• Hal showed a human quality when playing chess.
• Hal played trappy chess
• Meaning the obvious move for the opponent is a
  one of loss.
• Hal chose a slower strategy than current
  computers.
• Showing a more human side

8
Players Ratings

• How will our players match up
• Deep Thought 1987
• Deep Blue 1997
• Mac Hack 3
• Chess 3
• Chess 4
• Belle

9
Algorthims Trappy vs Minimax

• The term Trappy comes from the strategy of
  trying to trap your opponent into making a bad
  move.
• This algorithm needs creativity
• Minimax goes on the best possible move, assuming
  the opponent has chosen the best move.
• This algorithm goes the fast mate

10
Search Trees

• Games can be viewed as as search problems, then
  you can model it into a tree.
• The basic problem with looking at games as search
  trees is that these trees can get very large.

11
Strategy The Minimax Algorithm

• MiniMax Rules state optimal position as
  combination of best position for player and worst
  position for opponent
• Heuristic Evaluation Function assigns numerical
  value to each searched position
• Using Alpha-Beta Pruning with the heuristic
  values allows for
• eliminates repetitious evaluations
• provides a valid refutation for bad positions
• provides search tree in best-first order

heuristic evaluation minimax rules gt minimaxed
heuristic values
12
Computer Chess Masters
Players That Squared off to be come our champions
of this Artificial Chess Universe Chessmaster
8,000 Star Wars Chess Chessterfield vi3
13
The process by which we locked these Chess
Masters in battle for supremacy in our
Artificial Chess Universe

• Computer A
• CPU White
• Human Black
• Computer B
• CPU Black
• Human White
• Move
• CPU A Human B
• Human A CPU B

14
Chessterfield vi3
Representing our early Chess games

• Player Facts
• Product built by M. Luscher in his spare time
• Produced in 1999
• The .exe files size is 640KB and has no support
  files

15
Star Wars Chess

• Very Entertaining with the elements of your
  favorite Star Wars characters
• Over 72 different capture sequences
• Identifies many common defense as they are played

• Player Facts
• Product made for production built by Software
  Toolworks
• Produced in 1994
• Has the ability to play at various levels
• Beginner to Expert

16
Chessmaster 8,000

• Player Facts
• Product made for production built by Ubisoft
• Produced in 2001
• Has the ability to play with the tendencies of
  both current and past players

• Can be configured to play at any level and will
  rate you as a player
• LOTS of bells and Whistles (you get what you pay
  for)

17
Players and their Algorithmic Strategies Search depth Level of play
Chessterfield used minimax, evaluation function, cut-off test 5 ply novice
Star Wars Chess minimax and alpha-beta search 10 ply expert
ChessMaster 8,000 minimax, alpha-beta search, plus additional pruning, database of openings and end games 14 ply grand master
18
Results of Trials

• Despite the production years Star Wars (1994)
  chess had a more efficient search algorithm. It
  proved this by beating the Chessterfield (1999) 3
  matches in a row.
• However Star Wars could not beat ChessMaster
  8,000. Due less advanced pruning techniques

19
More Results

• Giving the different programs varying search
  times showed how the program searched the tree
• Given 30 sec. Chessterfield took between 5-30sec
  and made its move as soon as it found
  one--considered safe or if the time was up just
  made its safest known move
• Given 30 sec. Star Wars would take the whole 30
  sec. to make a move.
• Therefore we can see the Chessterfield used
  minimax moving when the best move was found. Star
  Wars used breadth first search to look through
  the entire level of the tree.

20
This Proves

• Computers became more advanced
• Better search algorithms
• Better chess techniques
• Therefore better chess players from computers.

21
Conclusion

• Computers have come a long way since the 1950s.
  From MacHack 3 to Deep Blue
• Theories and strategies behind computer chess
  have become more developed.
• The more efficient searches have proven advanced
  enough to outsmart some of the worlds greatest
  master.
• This was seen as the older verison (Chessterfield
  vi3) could not beat newer version (Star Wars)
  which could not beat the current grand master
  (Chessmaster 8,000.)

22
Why Computers Chess will not play on the same
plane as humans

• They are not able use psychology
• Psychology, as any top chess player will tell
  you, is an important key to winning chess. But
  Deep Blue has no psychological perception, can
  neither intimidate nor be intimidated, and
  experiences no joy from winning nor sadness from
  losing.
• However Deep Blue will use timing functions
• Given a total of 3.5 hours to make all its
  moves, it can ration time in a variety of ways.
  It can average the number of moves and attempt to
  deviate from that only by a small margin. Or it
  can move very fast, forcing Kasparov to respond.
  Or it can take an inordinate amount of time over
  one move, calculating many trillions of possible
  games, forcing Kasparov to wait and possibly
  become bored or agitated.

According to Research.IBM.com
23
Future of Computer Chess

• The only lacking element to make the perfect
  computer chess player is the ability to reason
• That means even now with the computers
  computation power, the computer cannot adapt to
  weaknesses in its opponents. The computer only
  knows the possible steps or the code can be
  manipulated to for the desired outcome
  (simulating trappy strategies)
• Once the program can recognize players game
  tendencies then it should adjust its game tree
  and heuristic value accordingly. Making an
  adaptable program.
• Deep Blue can force a win or a draw no matter
  what. Once the board is reduced down to 6 pieces
  in as little as 262 Moves.
• However chess tournaments only allows for 50
  moves. So Deep Blue making a brute force search
  with his algorithms and search trees wont do him
  much good in tournament style play.

24
THE END
OR
How About another Game DAVE?
25
References

• Hals Legacy published by MIT Press Cambridge,
  Massachusetts 1997
• Info About Hals play and history
• Reference.IBM.com
• Information about Deep Blue
• ChessBase.com
• Basic chess info and intresting facts
• www.informatik.uni-stuttgart.de/ipvr/ise/lehre/Pro
  blemSolving/
• Website about search methods
• www.tim-mann.org/chess.htm
• Info about deep thought
• www.chessmasternetwork.com
• Info about ChessMaster 8,000
• home.datacomm.ch/m.luescher
• Home of Chessterfield